Multi-agent reinforcement learning for autonomous vehicles: a survey

Abstract In the near future, autonomous vehicles (AVs) may cohabit with human drivers in mixed traffic. This cohabitation raises serious challenges, both terms of traffic flow and individual mobility, as well from road safety point view. Mixed fail to fulfill expected security requirements due heterogeneity unpredictability drivers, cars could then monopolize Using multi-agent reinforcement learning (MARL) algorithms, researchers have attempted design for scenarios, this paper investigates their recent advances. We focus on articles tackling decision-making problems identify four paradigms. While some authors address or without social-desirable AVs, others tackle case fully-autonomous latter is essentially a communication problem, most addressing admit limitations. The current driver models found literature are too simplistic since they do not cover drivers’ behaviors. As result, generalize over wide range possible For each investigated, we analyze how formulated MARL problem observation, action, rewards match paradigm apply.